Conversation AI · for research

Someone to talk to.
Something to learn from.

CogBot is an on-demand AI voice companion for seniors. For the people who study the conversation, it's a full research workbench — reach users, shape personas, refine prompts against simulated dialogue, and keep every call as data.

Explore the platform Live at cogbot.org · built for research teams

See the call, the way a senior does

A notification invites them to talk. One tap answers it, and they're in a warm, hands-free conversation — then a simple tap to hang up.

9:41

Thursday, June 19

CogBot

Jiho would love to talk

1 · Notification
Incoming call

Jiho

AI companion

2 · Incoming
Jiho

02:14

3 · In call
Call ended

02:38

4 · Ended

Natural voice, and a memory that grows

Seniors simply talk, whenever they like, and CogBot talks back — warmly and naturally. Every call adds to what it remembers, so the companion grows more personal over time.

Remembered across calls

Loves talking about her granddaughter

Knee pain — walking less these days

Enjoys her favorite old songs

carried into every future call

The conversation is the surface.
The platform is the point.

Behind the call screen sits an admin dashboard built for running a study — reaching users, engineering the model, and turning talk into analyzable data.

Reach users — one tap to a call

From the dashboard, send notifications: dialogue invitations or task reminders. Tapping one opens a VoIP-style incoming-call screen, so a gentle reminder turns straight into a real conversation. Reminders can be scheduled to go out on their own.

Admin dashboard

select users · compose a message

send a notification

CogBot · Conversation invite

Jiho would love to talk with you

now

tap

Incoming call → conversation starts

VoIP-style call screen

Tune prompts against simulated dialogue

Designing a good companion prompt is usually manual trial and error. CogBot makes it a workflow — define personas, scenarios, and evaluation criteria, then generate simulated conversations between CogBot and each persona. Revise the prompt automatically from those runs, or by hand. Personas give feedback too, and the same loop sharpens the personas themselves.

Two-stage loop — build realistic personas first, then refine the prompt against them.

Conversation simulation AI revise

CogBot

Have you had lunch yet today?

Persona · Min-su

Not yet — no appetite, really.

CogBot

I understand. Even a little would help keep your strength up.

Empathy Safety

Revise prompt

Every conversation becomes data

Transcripts and audio recordings are written automatically and synced to your own cloud storage — Google Cloud today, swappable to wherever your team works. Everything is ready for analysis the moment a call ends.

Conversation log Your cloud · synced

Jun 18 TXT · AUDIO

Jun 17 TXT · AUDIO

auto-written · ready for analysis

Refinement, as a loop you can actually run

The same machinery refines both sides. First it builds realistic personas; then it uses them to refine the prompt — over and over, with evaluation baked in.

STEP 1

Personas

realistic target users
STEP 2

Scenarios

topics & first turns
STEP 3

Simulated dialogue

CogBot × persona
STEP 4

Evaluate & revise

auto or by hand

loops back — sharpening personas, then the prompt

Actively in development

CogBot ships continuously. A few of the things on the bench right now.

IN BUILD

Native mobile apps

iOS & Android clients to replace the current web app.

PLANNED

Automated invitations

CogBot reaches out on its own — rule-based first, personalized next.